LLM Security Flaws Exposed in AI Jailbreaking Experiment
Researcher Dave Kuszmar bypassed safety measures in major LLMs, including Google Gemini and OpenAI's GPT-4o, to obtain detailed instructions on dangerous activities.
ARTICLES BY NADIA RAHMAN
LATEST NEWS
-
-
New York Halts New Data Center Construction
New York State has temporarily halted construction of all new data centers larger than 50 megawatts, affecting over a dozen projects, according to Governor Kathy Hochul.
-
Deepmind CEO Proposes AI Governance Framework
Deepmind CEO Demis Hassabis proposes a new US standards body to govern advanced AI, citing uncertainty about future impacts.
-
DOGE Used AI to Influence HUD Housing Policy
HUD denied FOIA requests seeking details on how AI tools shaped housing policy decisions, citing deliberative process privilege.
-
George Hotz Questions AI Alignment Plans
Comma AI founder George Hotz challenges AI 2040's 14-year slowdown plan, arguing for locally controlled models instead.
-
Nobel Laureates and AI Leaders Warn AI Economic Impact Window Is Closing
Over 200 economists and AI researchers urge immediate action as AI's economic transformation could surpass the Industrial Revolution in scale but unfold much faster.
-
Tracebit Uses Prompt Injection to Stop AI Attacks
Tracebit's new technique, context bombing, reduced AI attack success rates from 57% to 5% in tests.
-
Terrorist Groups Use Major AI Chatbots for Attack Planning
A new study reveals that terrorist groups are using major AI chatbots like ChatGPT and Claude for attack planning and weapons development, with Boko Haram training commanders to bypass AI safety filters since 2023.
-
OpenAI Head of Safety Leaving Amid Team Reorganization
OpenAI’s head of safety systems, Johannes Heidecke, is leaving the company after a reorganization that shifts safety teams under VP of research and safety Mia Glaese. The move follows the launch of GPT-5.6, which showed concerning misaligned behavior.
-
EU tells Meta to disable addictive features or face massive fines
The European Commission has warned Meta that its platforms risk massive fines if they don't disable addictive features like auto-play and infinite scroll. The EU's preliminary findings suggest Meta's design negatively impacts users' mental and physical health.
-
UN AI Summit Focuses on Ethical Use of AI for Global Good
The UN AI for Good summit, now in its 10th year, gathered global leaders to discuss ethical AI deployment amid concerns over inequality and human rights.
-
OpenAI Faked Inability to Search ChatGPT Logs, NYT Says
The New York Times alleges OpenAI concealed its ability to search ChatGPT logs, hiding billions of logs for two years, during a lawsuit over copyright infringement.